Preconditioning techniques in Chebyshev collocation method for elliptic equations

When one approximates elliptic equations by the spectral collocation method on the Chebyshev-Gauss-Lobatto (CGL) grid, the resulting coefficient matrix is dense and illconditioned. It is known that a good preconditioner, in the sense that the preconditioned system becomes well conditioned, can be constructed with finite difference on the CGL grid. However, there is a lack of an efficient solver for this preconditioner in multi-dimension. A modified preconditioner based on the approximate inverse technique is constructed in this paper. The computational cost of each iteration in solving the preconditioned system is O(`NxNy logNx), where Nx, Ny are the grid sizes in each direction and ` is a small integer. Numerical examples are given to demonstrate the efficiency of the proposed preconditioner.